Research KB 登录

PPR-1503.03585 Paper

Deep Unsupervised Learning using Nonequilibrium Thermodynamics

id
updated
type paper
title Deep Unsupervised Learning using Nonequilibrium Thermodynamics
authors Jascha Narain Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, S. Ganguli
venue ICML 2015
arxiv 1503.03585
tier 0
lifecycle EXTRACTED
epistemic n/a
ingested 2026-09-03
version arXiv:1503.03585
source-hash sha256:0000000000000000000000000000000000000000000000000000000000000000
admitted-under A2-direct-edge
admission-note G2DP 引用本工作提供扩散的非平衡热力学视角(Sohl-Dickstein et al., ICML 2015),支持把去噪理解为沿能量 landscape 的自由能下降。
citation-count-s2 10619

以非平衡热力学解释深度无监督学习(Sohl-Dickstein et al., ICML 2015):把训练刻画为自由能下降的渐进去噪,沟通统计物理与学习动力学。

与 G2DP 的关系(PPR-2606.26017:G2DP 引用本工作提供扩散的非平衡热力学视角(Sohl-Dickstein et al., ICML 2015),支持把去噪理解为沿能量 landscape 的自由能下降。

关联(1)

  • PPR-2606.26017 G2DP: Diffusion Planning with Spatio-Temporal Grid Guidance